OpenLane-3D车道数据集

本数据集面向自动驾驶环境感知算法研发团队、高精地图构建机构及ADAS系统开发商，旨在解决3D车道检测任务中真实场景数据稀缺、车道与路径对象关联标注缺失等瓶颈。数据集包含1,000个路段场景，拥有200K帧图像及超过880K条精细标注的车道线，并提供最近路径对象（CIPO）注释。与传统仅标注2D车道线的数据集不同，本数据集在3D空间中对车道几何进行建模，并将车道与可行驶路径对象关联，使模型能够理解“该车道通往何方向”等驾驶语义。This dataset is targeted at autonomous driving environment perception algorithm R&D teams, high-precision map construction institutions, and ADAS system developers, aiming to address bottlenecks in 3D lane detection tasks, including the scarcity of real-world scene data and the lack of associated annotations for lanes and path objects. The dataset contains 1,000 road segment scenes, with 200K frames of images and over 880K meticulously annotated lane lines, and provides Closest In-Path Object (CIPO) annotations. Unlike traditional datasets that only annotate 2D lane lines, this dataset models lane geometry in 3D space and associates lanes with drivable path objects, enabling models to understand driving semantics such as "which direction this lane leads to".

搜集汇总

背景与挑战

背景概述

OpenLane-3D车道数据集是一个面向自动驾驶环境感知的3D车道检测数据集，包含1,000个路段场景、200K帧图像和超过880K条精细标注的车道线。该数据集在3D空间中对车道几何进行建模，并提供最近路径对象（CIPO）注释，将车道与可行驶路径对象关联，以支持驾驶语义理解。